Threads

A thread is one piece of work, from the first prompt to done. It isn't a chat tab, and it isn't a branch. The agent's session is the thread, so there's no sidecar document to keep in sync, and starting one writes nothing into your repo.

WorkersCopy Anchor Link

The agent driving a thread is its worker. It's a real Claude Code, Codex or ACP agent process, running as you, with your credentials and your CLI configuration. Your skills, settings and CLAUDE.md or AGENTS.md apply just as they would in a session you started yourself.

A worker keeps going until it hits something only you can settle: a product call, a fork where a wrong guess would be expensive, or an irreversible action. Then it hands you an answerable question instead of a wall of text.

They run without the tabCopy Anchor Link

Every worker runs in its own detached background process, independent of the browser and of Ntwine itself. You can close the tab, quit the browser, stop the server or reboot, and your threads will still be there when you come back. Ntwine reconnects to the ones that are still running rather than replaying them from disk.

While an agent is working, Ntwine keeps the machine from idle-sleeping, so an overnight run doesn't stall because the laptop decided nobody was around. Closing the lid or pressing sleep still works.

Sub-agentsCopy Anchor Link

A worker can dispatch sub-agents for independent parts of its task. Ntwine binds each one back to its parent, so the fan-out is listed under the parent on its board and on its card. A parent waiting on them wakes as each one returns, and you can snooze a queued parent until all of them are back.

Names and status linesCopy Anchor Link

Each new thread gets a short name, its handle, which never changes once it's shown. While a thread works, a one-line status says what it's doing. Point at a thread's row to see it, or turn on Settings → Always show status lines to show it after every working thread's name. It also shows in the thread's own header.

Both are written by a quick model on your Claude sign-in. If you'd rather not, Settings → Background summaries turns them off. Threads are then named from your prompt or by their agent, with no status line.

Scratch directoryCopy Anchor Link

Each thread gets its own folder in the project, .ntwine/threads/<id>/, where its agent can keep notes and files. It stays empty unless the agent writes something.

Thread infoCopy Anchor Link

A thread's ⋯ menu has Thread info, which shows its tokens, turns, model requests, tool calls and cost. The same menu opens the thread in your editor, in a terminal or fullscreen, and deletes it.

Pinning and deletingCopy Anchor Link

Pin a thread to keep it at the very top of the list. You can delete a thread you no longer need from its menu, or have Ntwine delete done threads you haven't touched in a while under Settings. Pinned threads are never deleted.